Underwater acoustic target identification technology is one of the most advanced topics in the field of underwater acoustic research,which plays a potential role in military and civil fields in the future.Therefore,the related work of underwater acoustic target identification technology has certain application value.Traditional machine learning methods need to carry out artificial feature design.In order to overcome the drawbacks and defects of traditional methods,this paper proposes an underwater acoustic target recognition method based on deep learning,which takes advantage of self-optimization ability and strong adaptability of deep learning to classify and recognize underwater acoustic signals.Firstly,the underwater acoustic signal data is preprocessed to reduce the difficulty of extracting underwater acoustic signal features by removing blank segments,denoising and increasing,etc.The original data set is segmented to increase the amount of data,and a universal underwater acoustic signal data set suitable for deep learning network is constructed.Secondly,the influence of feature fusion on signal recognition accuracy is studied.In order to extract the meir spectrum and the cepstrum coefficient of Meir spectrum commonly used in acoustic target recognition technology from underwater acoustic signal,the frequency distribution and frequency component relationship of underwater acoustic signal are studied by using spectral contrast,chromatogram and tonality network.By comparing with the original signal as the input,the necessity of feature extraction is verified,and the accuracy of target recognition in the case of single feature input and fusion feature input is compared and analyzed.Third,to improve the recognition of traditional network,the convolutional neural network and the length of the memory for serial fusion network,on this basis,the research network recognition effect of different input case,verify the universality of the network,a single network and joint network for the classification of underwater acoustic signal recognition effect is analyzed.The research shows that the feature fusion proposed in this paper can extract the features of underwater acoustic signals comprehensively and effectively,and the joint network adopted in this paper has a great improvement in each recognition index compared with the traditional network,and the network has a strong robustness. |